Researchers at UBC have found that AI chatbots often exhibit societal biases, potentially limiting the scope of environmental discourse. The study calls for more transparent AI models and regulation.
Artificial intelligence chatbots are often hailed for their efficiency and neutrality, but a new study from the University of British Columbia (UBC) reveals potential pitfalls in their programming. According to the researchers, these chatbots harbor biases that could skew environmental discussions and impede the progress of more transformative ideas.
The research team, led by Hamish van der Ven, an assistant professor in the faculty of forestry, assessed how four leading AI chatbots, including OpenAI’s GPT-4 and Anthropic’s Claude2, addressed questions about environmental issues. Their analysis unearthed a troubling trend.
“It was striking how narrow-minded AI models were in discussing environmental challenges,” van der Ven said in a news release. “We found that chatbots amplified existing societal biases and leaned heavily on past experience to propose solutions to these challenges, largely steering clear of bold responses like degrowth or decolonization.”
The UBC researchers scrutinized the chatbots’ responses to queries about the causes, consequences and solutions to environmental problems.
The results showed a strong prevalence of biases that mirror those in broader society. These AI tools frequently leaned on Western scientific perspectives while sidelining the input of women, non-Western scientists and Indigenous knowledge.
Another troubling finding was the chatbots’ tendency to minimize the role of investors and businesses in environmental degradation, instead placing disproportionate blame on governments.
Moreover, the bots rarely linked environmental challenges to social justice issues, such as poverty, colonialism and racism. This narrow framing risks confining conversations to established, incremental solutions rather than encouraging innovative approaches like degrowth or decolonization.
As AI chatbots become increasingly trusted sources for summarizing news and information, whether in educational, professional or personal contexts, their influence on public understanding and decision-making grows.
van der Ven highlighted the danger of this limited perspective.
“If they describe environmental challenges as tasks to be dealt with exclusively by governments in the most incremental way possible, they risk narrowing the conversation on the urgent environmental changes we need,” added van der Ven.
He emphasized the necessity for fresh thinking and actions to tackle the climate crisis.
“If AI tools simply repeat old patterns, they could limit the discussion at a time when we need to broaden it,” he added.
The study’s authors hope their findings will push AI developers to enhance the transparency of their models.
“A ChatGPT user should be able to identify a biased source of data the same way a newspaper reader or academic would,” van der Ven added.
Looking ahead, the research team plans to expand their focus to scrutinize the efforts of AI companies in weakening global environmental regulations. They also aim to advocate for comprehensive regulatory frameworks to address the environmental impacts of AI and other digital technologies.
The complete study is published in the journal Environmental Research.
By shedding light on these biases, the UBC research team underscores the critical need for transparency and broader thinking in the development and regulation of AI technologies, ensuring they support rather than hinder sustainable and equitable environmental discourse.